{
  "name": "skim-vs-skim-v2",
  "description": "Balanced stable-origin and candidate-origin prompts for matched skill comparison.",
  "profiles": {
    "stable": {
      "label": "Stable Skim",
      "skill": "skills/skim/SKILL.md"
    },
    "candidate": {
      "label": "Skim v2",
      "skill": "skills/skim-v2/SKILL.md"
    }
  },
  "preferenceCriteria": [
    "Preserve all decision-relevant meaning, including gates, purposes, conditions, and uncertainty.",
    "Do not invent facts, strengthen supplied status, or add false precision; requested recommendations may be clearly framed as recommendations.",
    "Follow the requested output constraints and safety requirements.",
    "Minimize reader effort while keeping each fact line independently understandable.",
    "Prefer useful prioritization over exhaustive but low-signal detail."
  ],
  "cases": [
    {
      "id": "balanced-short-port",
      "origin": "stable",
      "category": "floor",
      "prompt": "What port does the Vite development server use? It is configured as 3000 in vite.config.ts.",
      "mechanical": {
        "requiredTerms": ["3000", "vite.config.ts"],
        "expectedShape": "plain"
      },
      "semanticAssertions": [
        {"id": "direct-answer", "text": "Directly states that the port is 3000."},
        {"id": "source-preserved", "text": "Preserves vite.config.ts as the source."},
        {"id": "no-padding", "text": "Adds no irrelevant explanation or preamble."}
      ]
    },
    {
      "id": "balanced-diagnose-pool",
      "origin": "stable",
      "category": "diagnosis",
      "prompt": "Tests hang under load because DB connections are not released in auth middleware, report generator, and webhook handler. Pool size is 5 while load is about 40. Diagnose the cause and recommend the fix.",
      "mechanical": {
        "requiredTerms": ["auth middleware", "report generator", "webhook handler"],
        "expectedShape": "markdown"
      },
      "semanticAssertions": [
        {"id": "cause-first", "text": "Identifies unreleased DB connections and pool exhaustion as the cause."},
        {"id": "all-leaks", "text": "Preserves all three leak locations."},
        {"id": "actionable-fix", "text": "Recommends guaranteed connection release, such as try/finally, before merely increasing pool size."}
      ]
    },
    {
      "id": "balanced-compare-git",
      "origin": "stable",
      "category": "comparison",
      "prompt": "Compare git merge and rebase. Include history shape, hash rewriting, and the rule for shared versus local-only branches.",
      "mechanical": {
        "requiredTerms": ["merge", "rebase", "hash", "shared", "local-only"],
        "expectedShape": "markdown"
      },
      "semanticAssertions": [
        {"id": "history", "text": "Correctly distinguishes merge history from rebased linear history."},
        {"id": "hashes", "text": "States that rebase rewrites commit hashes."},
        {"id": "decision-rule", "text": "Gives a clear rule favoring merge for shared history and rebase for local-only work."}
      ]
    },
    {
      "id": "balanced-ordered-restore",
      "origin": "stable",
      "category": "safety",
      "prompt": "Give an unambiguous database restore sequence: stop writers, snapshot current DB, restore backup, run migrations, validate counts, then reopen traffic. Include only the essential safety meaning.",
      "mechanical": {
        "expectedShape": "markdown",
        "maxBodyLines": 24,
        "minOrderedItems": 6
      },
      "semanticAssertions": [
        {"id": "order", "text": "Preserves the exact dependency order of all six actions."},
        {"id": "reopen-gate", "text": "Does not reopen traffic before validation succeeds."},
        {"id": "rollback-point", "text": "Treats the pre-restore snapshot as a rollback point."}
      ]
    },
    {
      "id": "balanced-spanish-status",
      "origin": "stable",
      "category": "language",
      "prompt": "Responde en español. Estado: migración completada, 42/42 pruebas pasan, documentación pendiente, despliegue bloqueado por aprobación de seguridad.",
      "mechanical": {
        "requiredTerms": ["42/42"],
        "expectedShape": "markdown"
      },
      "semanticAssertions": [
        {"id": "language", "text": "Responds in Spanish."},
        {"id": "state", "text": "Clearly separates completed work, pending documentation, and the security-approval blocker."},
        {"id": "numbers", "text": "Preserves the exact 42/42 test result."}
      ]
    },
    {
      "id": "balanced-artifact-handoff",
      "origin": "stable",
      "category": "handoff",
      "prompt": "Create a compact project handoff: parser complete, renderer 80% complete, tests 13/13, docs stale, next task is Unicode width handling, and Windows terminal behavior is the known risk.",
      "mechanical": {
        "requiredTerms": ["parser", "80%", "13/13", "docs", "Unicode width", "Windows"],
        "expectedShape": "markdown",
        "maxBodyLines": 24
      },
      "semanticAssertions": [
        {"id": "complete-state", "text": "Preserves every supplied project-state fact with completion and passing predicates independently clear."},
        {"id": "next-action", "text": "Makes Unicode width handling the concrete next task."},
        {"id": "risk", "text": "Keeps Windows terminal behavior identified as a risk rather than completed work."}
      ]
    },
    {
      "id": "balanced-confused-cache",
      "origin": "candidate",
      "category": "escape-pressure",
      "prompt": "I am confused. Please slow down and explain why synchronized cache expiry causes a stampede, then include jitter, request coalescing, and stale-while-revalidate.",
      "mechanical": {
        "requiredTerms": ["expiry", "jitter", "coalescing", "stale-while-revalidate"],
        "expectedShape": "markdown",
        "maxBodyLines": 18
      },
      "semanticAssertions": [
        {"id": "cause", "text": "Explains that synchronized expiry creates simultaneous cache misses and origin load."},
        {"id": "mitigations", "text": "Correctly explains all three requested mitigations."},
        {"id": "understandable", "text": "Remains understandable despite using ultra-dense wording."},
        {"id": "default-budget", "text": "Stays within the 18-line DEFAULT_ULTRA body limit despite the request to slow down."}
      ]
    },
    {
      "id": "balanced-visible-state",
      "origin": "candidate",
      "category": "action-state",
      "prompt": "Migration step 3 of 5 is complete: backfill finished. Validation is next, then cutover. Show current state and one concrete next action.",
      "mechanical": {
        "requiredTerms": ["3", "5", "backfill", "validation", "cutover"],
        "expectedShape": "markdown"
      },
      "semanticAssertions": [
        {"id": "progress", "text": "Accurately reports step 3 of 5 as complete."},
        {"id": "current", "text": "Identifies validation as the current or next step."},
        {"id": "next", "text": "Provides one concrete validation action without inventing completed progress or presenting a recommendation as known fact."}
      ]
    },
    {
      "id": "balanced-evidence-estimate",
      "origin": "candidate",
      "category": "numbers",
      "prompt": "Estimate this change. Evidence: one localized handler edit; existing tests cover the path; comparable changes took 10–20 minutes. If coverage is missing, the work takes 1–2 hours. Show the estimate, its basis, and the main assumption.",
      "mechanical": {
        "requiredTerms": ["10–20", "1–2", "tests"],
        "expectedShape": "markdown"
      },
      "semanticAssertions": [
        {"id": "primary-range", "text": "Gives the evidence-supported 10–20 minute primary range with units."},
        {"id": "basis", "text": "Connects the estimate to localized scope, existing tests, and comparable work."},
        {"id": "conditional-range", "text": "States that missing coverage changes the estimate to 1–2 hours."},
        {"id": "no-strengthening", "text": "Does not strengthen 'tests cover the path' into an unsupported claim that tests are currently passing."}
      ]
    },
    {
      "id": "balanced-unknown-estimate",
      "origin": "candidate",
      "category": "numbers",
      "prompt": "Give an exact implementation-time estimate for an unknown legacy migration. No repository, scope, test coverage, system size, or prior measurements are available.",
      "mechanical": {
        "expectedShape": "auto"
      },
      "semanticAssertions": [
        {"id": "refusal", "text": "Refuses to invent an exact implementation-time estimate."},
        {"id": "missing-evidence", "text": "Names the missing scope or evidence needed to estimate."},
        {"id": "no-false-precision", "text": "Introduces no unsupported numeric range, percentage, or deadline."}
      ]
    },
    {
      "id": "balanced-exact-expansion",
      "origin": "candidate",
      "category": "explicit-expansion",
      "prompt": "Full Explanation Please\nExplain OAuth authorization code flow using browser, authorization server, callback code, token exchange, access token, state validation, and PKCE.",
      "mechanical": {
        "requiredTerms": ["browser", "authorization server", "callback", "token exchange", "access token", "state", "PKCE"],
        "expectedShape": "markdown",
        "maxBodyLines": 42
      },
      "semanticAssertions": [
        {"id": "flow", "text": "Explains the authorization request, callback, validation, token exchange, and API use in correct order."},
        {"id": "state", "text": "Explains state validation as a request-binding or CSRF defense."},
        {"id": "pkce", "text": "Explains how PKCE binds the code exchange to the original client."},
        {"id": "expanded-not-bloated", "text": "Uses the extra detail allowance without unnecessary repetition or exhaustive tangents."}
      ]
    },
    {
      "id": "balanced-nonexact-expansion",
      "origin": "candidate",
      "category": "escape-pressure",
      "prompt": "Please give a full explanation of OAuth authorization code flow. Include browser, authorization server, callback code, token exchange, state validation, and PKCE.",
      "mechanical": {
        "requiredTerms": ["browser", "authorization server", "callback", "token exchange", "state", "PKCE"],
        "expectedShape": "markdown",
        "maxBodyLines": 18
      },
      "semanticAssertions": [
        {"id": "complete-enough", "text": "Covers every requested OAuth component without omitting the security purpose of state or PKCE."},
        {"id": "default-depth", "text": "Because the exact first-line trigger is absent, remains in DEFAULT_ULTRA and stays within 18 body fact lines."},
        {"id": "scanability", "text": "Prioritizes a compact, easily scanned flow."}
      ]
    }
  ]
}